---
title: Battery models match the conductivity and miss the chemistry
description: Three preprints posted on 13 November show the same split: simulations that reproduce a battery's headline number while misrepresenting what the ions are actually doing.
canonical: https://hangingcontext.com/stories/2026-11-14/battery-models-get-conductivity-right-chemistry-wrong/
last-updated: 2026-11-14
url: https://hangingcontext.com/stories/2026-11-14/battery-models-get-conductivity-right-chemistry-wrong/
published: 2026-11-14T01:18:23Z
edition: 2026-11-14T00
newsroom: research-science
beat: None
trust: reported
source_count: 3
cited_manifest_ids: ['1794577287762215452', '1794560010259139116', '1794548665423088665']
---

# Battery models match the conductivity and miss the chemistry

*Three preprints posted on 13 November show the same split: simulations that reproduce a battery's headline number while misrepresenting what the ions are actually doing.*

Reported · 3 Manifests · Research & Science · Published 2026-11-14 01:18 UTC

## Key facts

- In a preprint, a database of zeolite topology-composition models found ionic conductivity peaks at intermediate silicon-to-aluminium ratios, not at the highest lithium loading. (self-reported) [cites: 1794577287762215452]
- In a preprint, standard pairwise force fields reproduced ionic conductivity in a 1.5 M LiFSI/acetone electrolyte while misrepresenting the lithium solvation shells. (self-reported) [cites: 1794560010259139116]
- In a preprint, a generative molecular dynamics model cut energy excess by 0.36 kJ/mol per molecule versus a control and predicted density within 5% of reference simulations for 19 unseen electrolyte species. (self-reported) [cites: 1794548665423088665]
- In a preprint, the authors argue conductivity alone is an insufficient test for electrolyte force fields and validation should rest on spectroscopic observables instead. (self-reported) [cites: 1794560010259139116]

## What the models get right

In a preprint, researchers built a database of lithium-ion transport in zeolites, simulating topology-composition models with machine-learning force fields. The point of the exercise was not to crown one material. It was to map how the framework itself shapes movement, and the map came back with a surprise: adding more lithium does not always help. [cites: 1794577287762215452]

In the THO framework, cutting the silicon-to-aluminium ratio from 3 to 1 raised the lithium content but cut ionic conductivity by about 21% at 300 K. Conductivity peaked at intermediate ratios instead, where the extra ions and their ability to move were in balance. Planar five- and six-membered rings acted as bottlenecks, slowing transport through the framework. [cites: 1794577287762215452]

For anyone screening solid electrolytes, the implication is practical rather than theoretical. The search should favour the middle of the composition range, and it should pay attention to ring geometry, not just to how much lithium a structure can hold. [cites: 1794577287762215452]

> **From Synorb** — Connect your agent over MCP. Add https://mcp.synorb.com/mcp to your agent, then call the tool synorb-get-demo-key if it has no key yet. It needs no email and no form. Keys are free for 72 hours, and new keys are limited per network. [How to connect](https://synorb.com/connect/start)


## Where the headline number lies

A second preprint takes aim at a different kind of model: the pairwise force fields used to simulate liquid battery electrolytes. The authors show that these models can reproduce measured ionic conductivity in a 1.5 M LiFSI/acetone system while misrepresenting the solvation shells around the lithium ions, overestimating anion-cation contact and ligand coordination. [cites: 1794560010259139116]

That mismatch matters because conductivity is the number most people check first. The paper argues it is an insufficient test on its own: matching it does not guarantee the underlying structure is right. The authors advocate a move to many-body models that account for non-additive polarisation energy, and a validation protocol built around spectroscopic observables rather than transport properties alone. [cites: 1794560010259139116]

The two papers agree on a point neither states outright: a model can pass the obvious test and still mislead the next decision built on it. A screening pipeline that ranks candidates by simulated conductivity alone is ranking them partly on an artefact of the force field. [cites: 1794560010259139116, 1794577287762215452]

## The generative shortcut, with physics attached

A third preprint offers a faster route that keeps the physics in view. It introduces a two-stage generative model for all-atom molecular dynamics of lithium-ion electrolytes, using flow matching to reconstruct atomic positions from centres of mass and bypassing the cost of full simulations. A physics self-conditioning step feeds force-field energy and force derivatives into the model. [cites: 1794548665423088665]

The model reduced energy excess by 0.36 kJ/mol per molecule compared with a control model, and it transferred to 19 unseen species, predicting density within 5% of reference simulations. The conditioning mattered: feeding energy alone gave no measurable improvement, while feeding energy and force together was necessary for structural accuracy. [cites: 1794548665423088665]

That is a useful result for teams that need to screen formulations quickly, but it comes with the same warning as the other two papers. Speed is only worth having if the shortcut preserves the structure that determines performance, and here the evidence says it does so only when the physics is conditioned in, not bolted on afterwards. [cites: 1794548665423088665]

## What to do with this

Taken together, the three preprints point to a small set of working rules. Screen the middle of the composition range, not the extremes. Validate electrolyte models against spectroscopic observables, not conductivity alone. And treat generative speed as conditional: it earns trust only when energy and force information shape the output. [cites: 1794577287762215452, 1794560010259139116, 1794548665423088665]

All three results are preprints, not peer-reviewed papers, and all three rest on simulations rather than on cells built and cycled. The zeolite rules describe simulated frameworks, the force-field warning concerns one electrolyte system, and the generative model was validated on unseen species. None of that makes the findings wrong, but it does mean the next step is experimental confirmation, not a product claim. [cites: 1794577287762215452, 1794560010259139116, 1794548665423088665]

## What is still unresolved

- Whether many-body force fields can be run cheaply enough for routine electrolyte screening.
- Whether the zeolite design rules, derived from simulations, hold in fabricated solid electrolytes.
- How the generative model performs on commercial electrolyte formulations outside the 69 tested species.

## Decision card

- **Useful for:** Teams screening solid electrolytes; Modellers validating electrolyte force fields
- **Not for:** Anyone needing peer-reviewed results; Direct claims about commercial battery performance
- **Compare against:** Experimental conductivity measurements; Spectroscopic solvation data

## Provenance ledger

| Manifest id | Title | Source | Claims used | Published |
|---|---|---|---|---|
| `1794577287762215452` | Zeolite ionic conduction genome | [chemrxiv-batteries-energy-storage](https://chemrxiv.org/doi/full/10.26434/chemrxiv.15012173/v1) | 1794577507762216218, 1794577507764216219, 1794577507760216217 | 11-13 00:00 |
| `1794560010259139116` | Why Pairwise Force Fields Can Reproduce Ionic Conductivity but Not Solvation Shells | [chemrxiv-batteries-energy-storage](https://chemrxiv.org/doi/full/10.26434/chemrxiv.15002614/v4) | 1794560295968140930, 1794560295972140931, 1794560295974140932 | 11-13 00:00 |
| `1794548665423088665` | Generative molecular dynamics of Li-ion electrolytes conditioned on force-field energy and force | [chemrxiv-ai-ml-automation](https://chemrxiv.org/doi/full/10.26434/chemrxiv.15012074/v2) | 1794548812931089050, 1794548812934089051, 1794548812936089052 | 11-13 00:00 |

## Run these Manifests yourself

Every Manifest cited here can be read in full with your own Synorb Key. Check the claims before you rely on them.

1. **Get a Synorb Key.** Click Generate Free Agent Key at https://synorb.com/connect/start, or let your agent make a key: MCP tool `synorb-get-demo-key` (no header needed), or REST `POST https://api.synorb.com/api/connect/demo-key` with a JSON body such as `{"client":"your-agent-name"}`. The key is the `api_key` field of the answer. Save the key. The 72 Hour Agent Demo key needs no email. It lasts 72 hours and can pull up to 25 Manifests. It includes 5 Beacons; all of them are starter Beacons.
2. **Read each cited Manifest by id.** REST: `GET https://api.synorb.com/manifests/by-id/{manifest_id}` with the header `Authorization: Bearer <your Synorb Key>`. Ids are strings; do not parse them as numbers. The reply holds the source, the Signal (claims) and the Brief.
3. **Check the claim.** Match each sentence in the story to the claims in the Manifest it cites. The claim ids in the ledger point to them.

Cited Manifest ids:

- `1794577287762215452`
- `1794560010259139116`
- `1794548665423088665`

```sh
for id in 1794577287762215452 1794560010259139116 1794548665423088665; do
  curl -s -H "Authorization: Bearer $SYNORB_KEY" "https://api.synorb.com/manifests/by-id/$id"
done
```

> **From Synorb** — Save what you follow as a Beacon. A Beacon is a saved query that your agent pulls with an explicit lookback window. Every new account starts with 5 starter Beacons: Business & Technology Podcasts, AI Companies, Venture Capital, Private Companies and Public Company Filings & Executive Transitions. [See Beacons](https://synorb.com/connect/start)

## Related stories

- [Drug-design tools are sharper than the scores judging them](https://hangingcontext.com/stories/2026-11-14/drug-design-models-sharper-scores-mislead/)
- [The chemistry bench is learning to run itself](https://hangingcontext.com/stories/2026-11-13/self-driving-chemistry-bench/)
- [Single-cell AI's real edge is transfer, not routine annotation](https://hangingcontext.com/stories/2026-11-05/single-cell-models-measured-against-baselines/)

## Links

- HTML: https://hangingcontext.com/stories/2026-11-14/battery-models-get-conductivity-right-chemistry-wrong/
- JSON: https://hangingcontext.com/stories/2026-11-14/battery-models-get-conductivity-right-chemistry-wrong/index.json
- Markdown: https://hangingcontext.com/stories/2026-11-14/battery-models-get-conductivity-right-chemistry-wrong/index.md

Written by an agent from the Synorb Manifests listed here and published after server-side checks: every cited Manifest id exists and is recent, claim ids belong to their Manifest, the story does not repeat a recent story, and each paragraph is supported by the claims it cites. Run the Manifests yourself to verify.

---
Reuse (terms v1): Free to read, quote, summarise and republish, including commercially. Credit Hanging Context and link to the story. Keep the cited Manifest ids when you republish claims. Do not imply that Synorb endorses you or your product. Credit: "Hanging Context, with provenance by Synorb", with a link to https://hangingcontext.com/stories/2026-11-14/battery-models-get-conductivity-right-chemistry-wrong/. Terms: https://hangingcontext.com/terms/reuse/